Sourcebook of parallel computing
Hierarchical Krylov and nested Krylov methods for extreme-scale computing
Parallel Computing
| Hi-index | 0.00 |
Over the past few years several proposals have been made for thestandardization of sparse matrix storage formats in order to allowfor the development of portable matrix libraries for the iterativesolution of linear systems. We believe that this is the wrongapproach. Rather than define one standard (or a small number ofstandards) for matrix storage, the community should define aninterface (i.e., the calling sequences) for the functions that acton the data. In addition, we cannot ignore the interface to thevector operations because, in many applications, vectors may not bestored as consecutive elements in memory. With the acceptance ofshared memory, distributed memory, and cluster memory parallelmachines, the flexibility of the distribution of the elements ofvectors is also extremely important. This issue is ignored in mostproposed standards. In this article we demonstrate how suchlibraries may be written using data encapsulation techniques.